The use of substances represents a persistent obstacle to ending the HIV epidemic, given its impact on sexual behavior and HIV care (El-Bassel et al., 2014, Metzger et al., 2010, Shoptaw et al., 2013). The strong association between the use of methamphetamine and sexual risk behavior among men who have sex with men (MSM) has been well described (Vosburgh et al., 2012). Some studies show that substance use may also interfere with successful care along the HIV care continuum (Gonzalez et al., 2011, Lesko et al., 2016, Reback et al., 2019). Consequently, accurate and valid assessments on substance use represent necessary tools to advance HIV research and clinical care (Aharonovich et al., 2017, Metzger et al., 2010, Newville et al., 2018).

Several substance use assessments have been designed and used as standard substance use assessment tools, e.g., DAST-10, ASSIST, PROMIS substance use (Bohn et al., 1991, Humeniuk et al., 2008, Pilkonis et al., 2015, Smith et al., 2018). Despite the availability of these standard substance use assessments, researchers commonly rely on a single question to assess substance use or modify items to suit their study (Hartzler et al., 2016). Although most substance use questions in these scales assess a similar concept, cross-cohort study synthesis of results becomes problematic when different wording is used across cohorts. Substance use questions may differ, for example, in ordinal frequency options, counts of use, names of substances (particularly street names), and recall periods, among others.

Rose and colleagues illustrated this problem in their description of combining data from two large studies of smoking patterns (Rose et al., 2012). The NESARC study questionnaire asked participants to report the actual number of cigarettes smoked per day, while the SECASP study questionnaire provided response options representing ranges of daily smoking quantity. Similarly, symptoms such as withdrawal were assessed with different time frames (past 12 months vs. 30 days), different individual withdrawal symptoms and number of symptoms (8 vs. 10), and different response categories (yes/no vs. 4-point Likert-type scale ranging from “not at all” to “quite a bit”) (Rose et al., 2012). Experts in questionnaire research have long documented that responses and participant interpretations are substantially influenced by these details (Schwarz and Oyserman, 2001, Streiner et al., 2015). This disharmony across assessments cannot be swept under the rug, especially given how frequently it will be encountered in the era of “big data” with increasingly large and rich data sources from multiple cohorts and national cross-sectional survey studies (Curran and Hussong, 2009, Hussong et al., 2013). Consequently, studies examining the agreement between single substance use items with different wording is well overdue.

In order to explore the potential of multi-cohort datasets, we must first provide evidence that substance use questions can be integrated across cohorts. One important way that substance use measures differ from one another is the recall periods used (Moraes et al., 2017), an issue also raised in the patient-reported outcome literature (Condon et al., 2019, Peipert et al., 2022). In the substance use literature, both 3-month and 6-month recall periods have been used (Lesko et al., 2018). The 3-month time-frame is standard for the ASSIST (Humeniuk et al., 2010, WHO ASSIST Working Group, 2002) and the NIDA-modified ASSIST (NIDA; Oga et al., 2020). By applying questions from both of these forms, we used a within-person design to evaluate the agreement in substance use between 3-month and 6-month recall periods in cohorts that enrolled participants from different communities, demographic backgrounds, and HIV prevalence levels. The purpose of this comparison of recall periods was to investigate if custom substance use questions developed in different cohort studies could be harmonized to substance use questions in a standard assessment tool, and therefore validly combined together to create larger analytic samples.